BBBClub: Bookbinder's Book Club
In evtree: Evolutionary Learning of Globally Optimal Trees

Description Usage Format Details Source References Examples

Marketing case study about a (fictitious) American book club to whose customers a book about “The Art History of Florence” was advertised.

1	data("BBBClub")

A data frame containing 1,300 observations on 11 variables.

choice: factor. Did the customer buy the advertised book?
gender: factor indicating gender.
amount: total amount of money spent at the BBB Club.
freq: number of books purchased at the BBB Club.
last: number of months since the last purchase.
first: number of months since the first purchase.
child: number of children's books purchased.
youth: number of youth books purchased.
cook: number of cookbooks purchased.
diy: number of do-it-yourself books purchased.
art: number of art books purchased.

The data is a marketing case study about a (fictitious) American book club. taken from the Marketing Engineering textbook of Lilien and Rangaswamy (2004). In this case study, a brochure of the book “The Art History of Florence” was sent to 20,000 customers and 1,806 of which bought the book. A subsample of 1,300 customers is provided in BBBClub for building a predictive model for choice.

The use of a cost matrix is suggested for this dataset. Classifying a customer that purchased the book as a non-buyer is worse (cost = 5), than it is to classify a custumer that did not purchase the book as a buyer (cost = 1).

Complements to Lilien and Rangaswamy (2004).

Lilien GL, Rangaswamy A (2004). Marketing Engineering: Computer-Assisted Marketing Analysis and Planning, 2nd edition. Victoria, BC: Trafford Publishing.

## Not run: 
## data, packages, random seed
data("BBBClub", package = "evtree")
library("rpart")
suppressWarnings(RNGversion("3.5.0"))
set.seed(1090)

## learn trees
ev <- evtree(choice ~ ., data = BBBClub, minbucket = 10, maxdepth = 2)
rp <- as.party(rpart(choice ~ ., data = BBBClub, minbucket = 10, model = TRUE))
ct <- ctree(choice ~ ., data = BBBClub, minbucket = 10, mincrit = 0.99)

## visualization
plot(ev)
plot(rp)
plot(ct)

## accuracy: misclassification rate
mc <- function(obj) 1 - mean(predict(obj) == BBBClub$choice)
c("evtree" = mc(ev), "rpart" = mc(rp), "ctree" = mc(ct))

## complexity: number of terminal nodes
c("evtree" = width(ev), "rpart" = width(rp), "ctree" = width(ct))

## compare structure of predictions
ftable(tab <- table(evtree = predict(ev), rpart  = predict(rp),
  ctree  = predict(ct), observed = BBBClub$choice))

## compare customer predictions only (absolute, proportion correct)
sapply(c("evtree", "rpart", "ctree"), function(nam) {
  mt <- margin.table(tab, c(match(nam, names(dimnames(tab))), 4))
  c(abs = as.vector(rowSums(mt))[2],
    rel = round(100 * prop.table(mt, 1)[2, 2], digits = 3))
})

## End(Not run)

Loading required package: partykit
Loading required package: grid
Loading required package: libcoin
Loading required package: mvtnorm
   evtree     rpart     ctree 
0.2430769 0.2384615 0.2476923 
evtree  rpart  ctree 
     4      5      8 
                   observed  no yes
evtree rpart ctree                 
no     no    no             799 223
             yes             38  24
       yes   no               0   0
             yes             12  18
yes    no    no               0   0
             yes              0   0
       yes   no              21  19
             yes             30 116
     evtree   rpart   ctree
abs 186.000 216.000 238.000
rel  72.581  70.833  66.387